Simon Willison’s Weblog is the densest real-time record of what software engineers can actually do with LLMs, published daily by someone who uses them in his own practice.
No other voice tracks the tool-and-model churn of agentic development day-by-day with the same density and the same willingness to name what is hype; if you follow one feed for this domain, this is the one.
What it is #
A long-running personal blog, plus a companion TIL site and a monthly sponsor-funded digest, written by Simon Willison, the Django co-creator and author of Datasette and the LLM CLI.
He writes first-person, every day, about models, coding agents, MCP, sandboxing, and Python tooling, usually from direct hands-on use rather than press releases.
He coined the widely quoted framing of LLMs as a “weird, over-confident intern” and maintains an Agentic Engineering Patterns stream that distills working practice.
Status #
Active and as influential as ever.
As of 2026-10-15 the homepage shows posts daily, sometimes several a day, through 14 October 2026, led in the Highlights list by the Tree Appreciator game post (2026-10-14, a Claude Opus 5.5 spot-the-difference game he prompted in the plain chat app and let crunch for three hours), with the newer items around it including 13 October’s coverage of Anthropic’s custom source-control system Scion and his llm-gemini 0.35 release adding the new Gemini 4 Argon model, and behind them “A new feature for my blog, built using my voice” (
2026-10-09, a 1,294-word first-person account of shipping a blog feature almost entirely by talking to Codex voice mode while cooking dinner, then finishing it with a typed PR review), the Deno-joining-Cloudflare note and a Matthew Green quote on the odds of losing confidence in public-key encryption (both 9 October), the Python 3.15 GitHub Actions note and the New York Times report that Anthropic agents submitted 20 incomplete visa applications through the State Department site (both 10 October), an 11 October note on Dwarf Fortress finally adopting version control, and 12 October items led by “Hello Python 3.15, goodbye 3.10, and now I can use 3.11” alongside his web-edit 0.1a0 and datasette-ports 0.4 releases, with the recent landmark essays “We’re going to need default hard budget caps on pretty much everything” (2026-10-03, his case for hard, default spending limits on pay-by-usage services now that agents can run up surprise bills), “2026 in LLMs (so far)” (2026-09-27, his 7,771-word annotated keynote from the WeAreDevelopers World Congress North America), the OpenAI DevDay 2026 live blog (2026-09-29, filed from the keynote at Fort Mason), and “Claude Opus 5.5, GPT-6 Sol, GPT-6 Luna, and a new price war” (2026-09-22), plus dedicated tag streams for agentic-engineering, coding-agents, and ai-assisted-programming.
Around the landmarks the security-and-cost thread keeps running: quotes on Meta’s Muse agent (2026-09-25 and 2026-09-28), the Anthropic red team’s GLM-5.3 cyber-capability report (2026-09-29), and Matthew Green’s sandboxing-versus-agent-worms essay (2026-10-01).
He is also convening offline: a Birds of a Feather session on agentic engineering with Jesse Vincent in San Francisco, announced 2026-09-23 for 14 October.
He ships code on the blog too: the sqlite-utils 4.0 release notes record that much of it was written by a coding agent, so the blog is documentation of his own agent loops, not just commentary on them.
Strengths #
- Daily cadence means he catches and explains changes while they are still news, not after they settle into a retrospective.
- He uses the tools he writes about: agent-written releases, sandboxed agent experiments, and prompt-injection testing all appear as first-hand notes.
- The tag archives (agentic-engineering, code-review, sandboxing) act as a searchable playbook, not just a feed.
- Repeatedly surfaces skeptical and security-critical coverage, from prompt-injection results to supply-chain attacks.
Cautions #
- Volume is high; without a filter the firehose buries the signal, which is why the sponsor digest exists.
- Coverage is Python-and-open-source weighted, so the IDE-centric and enterprise story gets less depth.
- He is a tool builder (LLM, Datasette) which biases the notes toward command-line, open, and local workflows.
Pricing #
Free to read. The blog is sponsor-funded (GitHub sponsors at $10/month buy a monthly curated digest), with no paywall on the core feed.
Price history #
| Date | Plan | Change | Source |
|---|---|---|---|
| 2026-09 | GitHub Sponsors | Baseline: blog free to read, sponsor-funded; $10/mo sponsorship buys the monthly curated digest. |
Compared to #
- Latent Space: the podcast-and-conference synthesis of the same landscape; read Willison for the day-to-day working notes, listen to Latent Space for the interviews and the industry view.
- Steve Yegge: the provocative thesis-setter versus the reliable daily chronicler; Yegge declares, Willison documents.
- The Pragmatic Engineer: engineering-org and industry analysis versus hands-on tool notes; Orosz for the org, Willison for the terminal.
Bottom line #
Recommended for any engineer who works with LLMs daily and wants a low-DSP, high-signal feed plus a searchable archive of working patterns. Not for people who prefer periodic synthesis over daily volume, or who want a vendor-neutral enterprise take.
Top 5 recommended reading #
Vibe engineering - the essay that names the disciplined end of AI-assisted programming and lists the senior-engineering skills it rewards.
Changes #
- 2026-08-29 - Created in the People and publications category seed.
- 2026-09-16 - Posting recency refreshed: the homepage now shows daily posts through 15 September 2026 (previously recorded through 12 September).
- 2026-09-18 - Posting recency refreshed: the homepage now shows daily posts through 17 September 2026.
- 2026-09-20 - Added the Price history section tracking price changes in a table, per the new owner rule.
- 2026-09-22 - Posting recency refreshed: daily posts through 22 September 2026, led by the Opus 5.5/GPT-6 Sol/Luna price-war essay; he also announced a San Francisco agentic-engineering Birds of a Feather session with Jesse Vincent for 14 October.
- 2026-09-24 - Added the Top 5 recommended reading section.
- 2026-09-27 - Posting recency refreshed: daily posts through 26 September 2026.
- 2026-09-29 - Posting recency refreshed: daily posts through 28 September 2026, now led by the “2026 in LLMs (so far)” keynote writeup (2026-09-27) and Claude Sonnet 5.5 coverage (2026-09-28).
- 2026-09-30 - Posting recency refreshed: daily posts through 29 September 2026, led by the OpenAI DevDay 2026 live blog.
- 2026-10-04 - Posting recency refreshed through 3 October 2026, added the 3 October default-hard-budget-caps essay as the newest landmark with the Muse-to-sandboxing quote thread around it, and corrected the BoF announcement date to 2026-09-23.
- 2026-10-08 - Posting recency refreshed: daily posts through 7 October 2026, now led by “Claude Haiku 5.5” (2026-10-07) alongside the Mistral Large 4 and rogue-agents-on-Wikimedia items.
- 2026-10-10 - Posting recency refreshed: daily posts through 9 October 2026, now led by “A new feature for my blog, built using my voice” (2026-10-09, the voice-driven Codex build with a typed review at the end), added to Status and References (URL fetched this run).
- 2026-10-11 - Posting recency refreshed: daily posts through 11 October 2026, with the 9 October voice-built Newsletters page still leading the Highlights list and the newer items around it (the Deno-to-Cloudflare acquisition, the Matthew Green encryption quote, the Python 3.15 GitHub Actions note, the Anthropic-agents visa-applications report, and the 11 October Dwarf Fortress version-control note); homepage fetched this run.
- 2026-10-13 - Posting recency refreshed: daily posts through 12 October 2026, led by “Hello Python 3.15, goodbye 3.10, and now I can use 3.11” with the web-edit 0.1a0 and datasette-ports 0.4 releases around it; homepage and post URLs fetched this run.
- 2026-10-15 - Posting recency refreshed: daily posts through 14 October 2026, now led by the Tree Appreciator game post with 13 October’s Scion source-control coverage and the llm-gemini 0.35 release for Gemini 4 Argon around it; homepage and post URLs fetched this run.
See also #
- Agentic Coding Tools Landscape - the map his daily notes keep current
- Context Management Patterns - a pattern stream he has written into explicitly
- Model Selection for Coding Tasks - the model churn he reports on daily, distilled to a process
- The Codebase Gardener - the human role his hands-on notes presuppose
References #
https://til.simonwillison.net/ - the Today I Learned companion archive
https://llm.datasette.io/ - the LLM CLI he builds and documents on the blog
https://arstechnica.com/ai/2025/03/is-vibe-coding-with-ai-gnarly-or-reckless-maybe-some-of-both/ - a critical press take on vibe coding that quotes his views